The artificial-intelligence boom is entering a more consequential phase: investors are no longer asking only how quickly the technology can spread, but whether the companies building its infrastructure can earn enough to repay the money being borrowed to build it. That shift is unsettling markets and turning a technology story into a broader test of corporate finance, electricity demand and economic policy.
Recent market indicators point to growing concern over the sector’s financing model. Deloitte reported that credit-default-swap prices and yields on AI-related bonds have risen as investors question whether future revenue from AI companies will service the debts of firms constructing data centres and other infrastructure. Gartner has separately projected that 80% of “neocloud” providers could cease independent operations by 2030, largely because capital spending may outrun returns. Deloitte’s weekly global economic update links the pressure to a widening gap between investment and demonstrable cash flow.
The economics behind the anxiety
The central wager is straightforward but enormous. Cloud companies, chipmakers, utilities and data-centre developers are committing capital today in the expectation that demand for AI computing will remain strong for years. That demand depends on businesses adopting AI at scale, consumers paying for new services and model developers converting impressive technical performance into durable margins.
Those assumptions are not equally secure. Training and operating advanced models require expensive chips, specialised data centres and large amounts of electricity. The infrastructure providers may benefit even if one model company loses market share, but they remain exposed if customers reduce orders, renegotiate contracts or fail to generate enough revenue from AI applications. High interest rates make that exposure more serious by increasing the cost of projects whose profits may arrive only later.
The concern is therefore less a claim that AI lacks value than a question of timing and distribution. Productivity gains could eventually justify today’s investment, but the benefits may accrue to a small group of dominant firms while weaker infrastructure companies absorb the losses. A rapid expansion followed by consolidation would not necessarily end AI’s development; it could instead shift control toward companies with the strongest balance sheets.
Why markets are watching beyond technology
AI-related borrowing is becoming part of a wider debate about corporate debt. Deloitte also reported accelerating US inflation and noted that investors were increasingly worried about AI-linked credit. If lenders demand higher yields, new data-centre projects become more expensive. Developers may then delay construction, seek equity financing or pass costs to cloud customers and electricity users.
That transmission matters because the AI build-out is tied to physical infrastructure. Data centres compete for grid capacity, land and power-generation equipment. Communities may welcome construction and high-skilled jobs, while questioning whether public subsidies, water use and electricity costs are justified by uncertain long-term benefits. The more expensive financing becomes, the more pressure companies face to demonstrate that local economic gains will outlast the construction cycle.
There is also a geopolitical dimension. The current AI race is increasingly treated as strategic competition, encouraging governments to support domestic computing capacity and semiconductor supply chains. Public backing can reduce private-sector risk, but it can also transfer losses to taxpayers if forecasts prove too optimistic. Policymakers must balance technological competitiveness against competition concerns, energy constraints and financial stability.
Two plausible paths forward
In the optimistic scenario, falling computing costs and better-performing models unlock applications in software, medicine, manufacturing and research. Revenue catches up with infrastructure spending, bond spreads narrow and the sector undergoes a healthy shakeout rather than a systemic crisis. Smaller providers that offer specialised services could survive by serving profitable niches rather than competing directly with the largest platforms.
In the less favourable scenario, customers discover that many AI deployments do not produce sufficient savings or new income. Orders slow, highly leveraged providers struggle and lenders reprice the entire sector. A correction could damage technology shares and construction suppliers without eliminating demand for AI itself. The result would likely be fewer independent providers, more acquisitions and greater dependence on the largest cloud and model companies.
The next evidence will come from cash flow, not announcements. Investors will look for signed customer commitments, utilisation rates, debt-service coverage and measurable productivity gains. Governments will face pressure to disclose the public cost of energy and infrastructure support. Companies, meanwhile, will need to show that capital expenditure reflects real demand rather than a race to secure market position.
The AI boom is not over, but its easy narrative is. The technology’s future will depend less on the scale of promised breakthroughs than on whether those breakthroughs can support the financial structure being built around them.